use_direction_classifier=False causes AttributeError for NoneType

[filaPro]

Looks like use_direction_classifier=False bring us here and after that this None gets here. Where _loss must be a Tensor not None.

[Tai-Wang]

Thanks for your information. Actually, we have not tried to remove the direction classifier for long. The problem does exist from the first look. Do you encounter this bug in your experiments?
Maybe a workaround for now will be setting the default dir loss to 0*something. More elegant ways to avoid this should be adopted later. Or do you have any ideas?

[filaPro]

Yes, it's just about my own experiments. Your 0*something can be a solution.